Identifying anti-tumor heat shock proteins based on evolutionary information using deep learning method

Abstract

Heat shock proteins (HSPs) belong to stress proteins. The functions of HSPs are mainly reflected in three aspects: molecular chaperones, regulation of apoptosis and immune responses. Recent studies have shown that there is a certain correlation between HSPs and tumor cell. HSPs are participated in the invasion, proliferation and metastasis of tumor cells. Therefore, developing an accurate model for identification anti-tumor HSPs is a key step to understand molecular functions of HSPs and human tumor diseases. In this study, we propose using deep learning methods to identify anti-tumor HSPs. To seek out the optimal model for the dataset, several hyper-parameters are optimized according to the results of 10-fold cross-validation. Finally, the performance of the proposed model is further determined through an independent dataset. The experimental results indicated that the proposed model could classify anti-tumor HSPs with accuracy (ACC) of 93.76%, sensitivity (SN) of 92.80%, specificity (SP) of 93.33%, and Matthew's correlation coefficient (MCC) of 86.39% on the 10-fold cross-validation. Compared with other deep learing methods, using convolutional neural network (CNN) can achieve a significant improvement for identifying of anti-tumor HSPs.